Search

Speech enhancement is a demanding task in automated speech processing pipelines, focusing on separating clean speech from noisy channels. Transformer based models have recently bested RNN and CNN models in speech enhancement, however at the same time they are much more computationally expensive and require much more high quality training data, which is always hard to come by. In this paper, we present an improvement for speech enhancement models that maintains the expressiveness of self-attention while significantly reducing model complexity, which we have termed Spectrum Attention Fusion. We carefully construct a convolutional module to replace several self-attention layers in a speech Transformer, allowing the model to more efficiently fuse spectral features. Our proposed model is able to achieve comparable or better results against SOTA models but with significantly smaller parameters (0.58M) on the Voice Bank + DEMAND dataset.

Presence of multiple competing periodicities may result in a system to go through states with modulated periodicities, an example of which is the self-similar staircase-like structure called the Devil's staircase. Herein we report on a novel staircase structure of domain periodicity in an amorphous and achiral Fe/Gd magnetic thin film wherein the reciprocal space wavevector \textbf{Q} due to the ordered stripe domains does not evolve continuously, rather exhibits a staircase structure. Resonant X-ray scattering experiments show jumps in the periodicity of the stripe domains as a function of an external magnetic field. When resolved in components, the step change along Q$_x$ was found to be an integral multiple of a minimum step height of 7 nm, which resembles closely to the exchange length of the system. Modeling the magnetic texture in the Fe/Gd thin film as an achiral spin arrangement, we have been able to reproduce the steps in the magnetization using a Landau-Lifshitz spin dynamics calculation. Our results indicate that anisotropy and not the dipolar interaction is the dominant cause for the staircase pattern, thereby revealing the effect of achiral magnetism., Comment: 10 pages, 5 figures

This paper addresses the robustness of a network to sustain its connectivity and controllability against malicious attacks. This kind of network robustness is typically measured by the time-consuming attack simulation, which returns a sequence of values that record the remaining connectivity and controllability after a sequence of node- or edge-removal attacks. For improvement, this paper develops an efficient framework for network robustness prediction, the spatial pyramid pooling convolutional neural network (SPP-CNN). The new framework installs a spatial pyramid pooling layer between the convolutional and fully-connected layers, overcoming the common mismatch issue in the CNN-based prediction approaches and extending its generalizability. Extensive experiments are carried out by comparing SPP-CNN with three state-of-the-art robustness predictors, namely a CNN-based and two graph neural networks-based frameworks. Synthetic and real-world networks, both directed and undirected, are investigated. Experimental results demonstrate that the proposed SPP-CNN achieves better prediction performances and better generalizability to unknown datasets, with significantly lower time-consumption, than its counterparts., Comment: 10 pages, 7 figures, 14 pages Supplementary Information

We investigate the experimental signatures of Raman spectroscopy of bi- and tri-magnon excitations in the distorted triangular lattice antiferromagnets alpha-LCr2O4 (L=Sr, Ca). We utilize spin wave theory to analyze the nearly 120 degree spin-3/2 spiral ordered antiferromagnetic ground state to compute the single-magnon density of states, single-magnon dispersion, and bimagnon and trimagnon Raman spectra (polarized and unpolarized). It is found that Raman scattering is capable of capturing the effect of the rotonlike M and M' points on the bimagnon Raman spectrum. Our calculation confirms the connection between single-magnon rotonlike excitation energy and bimagnon Raman excitation spectrum observed experimentally. The roton energy minimum in momentum space is half of the energy of a bimagnon excitation signal. The experimental magnetic Raman scattering result displays two peaks which have a Raman shift of 15 meV and 40 meV, respectively. Theoretical modeling and analysis of the experimental spectrum of alpha-SrCr2O4 within our distorted Heisenberg Hamiltonian lattice suggests that the low-energy peak at 15 meV is associated with the bimagnon excitation, whereas the high-energy peak around 40 meV is primarily a trimagnon excitation. Based on our fitting procedure we propose a new set of magnetic interaction parameters for alpha-SrCr2O4. These parameters reproduce not only the experimental Raman spectrum, but also the inelastic neutron scattering response (including capturing high energy magnon branches). We also compute the unpolarized bimagnon and trimagnon Raman spectra for alpha-CaCr2O4. Furhtermore, we found that the polarization sensitivity of Raman spectrum can be utilized to distinguish the bi- and tri-magnon excitation channels., Comment: 11 pages, 4 figures

Connectivity and controllability of a complex network are two important issues that guarantee a networked system to function. Robustness of connectivity and controllability guarantees the system to function properly and stably under various malicious attacks. Evaluating network robustness using attack simulations is time consuming, while the convolutional neural network (CNN)-based prediction approach provides a cost-efficient method to approximate the network robustness. In this paper, we investigate the performance of CNN-based approaches for connectivity and controllability robustness prediction, when partial network information is missing, namely the adjacency matrix is incomplete. Extensive experimental studies are carried out. A threshold is explored that if a total amount of more than 7.29\% information is lost, the performance of CNN-based prediction will be significantly degenerated for all cases in the experiments. Two scenarios of missing edge representations are compared, 1) a missing edge is marked `no edge' in the input for prediction, and 2) a missing edge is denoted using a special marker of `unknown'. Experimental results reveal that the first representation is misleading to the CNN-based predictors., Comment: In Proceedings of the IEEE 2022 International Joint Conference on Neural Networks (IJCNN)